SHRP 2 Naturalistic Driving Study (NDS) Jim...

SHRP 2 Naturalistic Driving Study (NDS)

Jim HedlundHighway Safety North

Nebraska Highway Safety Conference

Lincoln, NE

March 21, 2017

Accelerating solutions for highway safety, renewal, reliability, and capacity

In the next 40 minutes

• What’s a Naturalistic Driving Study?

• SHRP 2 NDS study – high-level overview

• NDS analyses – examples of research topics and goals

• How to use the data

• NDS website demonstration

• Questions, discussion, where to go for more information

2

3

Naturalistic Driving Studies

• Method: instrument volunteer drivers’ vehicles – continuously

observe and record their driving for months or years

– What do drivers really do? Speeding, tailgating, cell phone, alcohol …

– What were they doing just before they crashed?

Usual crash studies can only guess.

– How do drivers react to cues from the vehicle and environment?

4

SHRP 2 Study Design

Two linked databases providing unprecedented detail on driver

behavior and driver interaction with roadway features

• Naturalistic Driving Study (NDS) data – VTTI, Virginia Tech

– record of every trip by volunteer drivers over 12-24 months

– 3,147 drivers, male and female, all ages

– 1,900 vehicles on the road at any time

• Roadway (RID) data – CTRE, Iowa State

– from mobile van data collection in study areas

– from roadway inventory

– from other state data sources

VT

TI

Drivin

g T

ransport

ation w

ith T

echnolo

gy

NDS Study Design

Largest Naturalistic Driving Study Ever Undertaken

• 3,147 drivers, all age/gender groups.

• 3,958 data years; 5 M trip files; 35 M vehicle miles

• 2 years of data collection

• Most participants 1 to 2 years

• Vehicle Types: All light vehicles

• Passenger Cars

• Minivans

• SUVs

• Pickup Trucks

• Six data collection sites

Integrated with detailed

roadway information

150

vehicles

300

vehicles

450

vehicles

5

VT

TI

Drivin

g T

ransport

ation w

ith T

echnolo

gy

NDS Data Overview

Driver demographics, assessments

Vehicle descriptors

TRIP DATA

Multiple Videos

Machine Vision

• Eyes Forward Monitor

• Lane Tracker

Accelerometer Data (3 axis)

Rate Sensors (3 axis)

GPS

• Latitude, Longitude, Elevation, Time, Velocity

Forward Radar

• X and Y positions

• X and Y Velocities

Cell Phone Records

• Beginning and end of all cell phone conversations on major carriers

Passive Alcohol Sensor

Illuminance sensor

Infrared illumination

Incident push button

• Audio (only on incident push button)

Turn signals

Vehicle network data

• Accelerator

• Brake pedal activation

• ABS

• Gear position

• Steering wheel angle

• Speed

• Horn

• Seat Belt Information

• Airbag deployment

• Many more variables…

6

Data Acquisition System (DAS)

7

NDS Data Example

8

9

Roadway (RID) Data Overview

• New data: collected at highway speed from van, 12,538 centerline

miles (both directions)

– curvature location, length, radius; grade; cross-slope; lane number,

width, type; shoulder type (width if paved); all MUTCD signs;

medians; barriers; rumble strips; lighting; intersection location,

number of approaches, and control type; videolog

• Existing data from ESRI and state inventories in 6 study states:

any available roadway information – varies by state

• Supplemental data from 6 states: traffic, weather; work zones;

crashes; roadway improvements; laws

Scope: 12,538 centerline miles in the 6 NDS sites

Method: Instrumented Van

Mobile Data

11

Route Name Direction Chainage State Collection Date

Videolog Images

Mobile Van Data

• New data SHRP 2 collected

• Quality assured to meet project

specs

• 25,000 driven/ 12,500 centerline

miles across the six NDS sites

All data (mobile van data and acquired data)

are referenced to a common basemap that

covers the continental US

Acquired Roadway Data

• Horizontal Curvature:

Radius, Length, PC, PT,

Direction

• Grade

• Cross Slope

• Lane in terms of the number,

width, and type ( turn, passing,

acceleration, car pool, etc…)

• Shoulder type/curb; paved

width if exists

• Intersection location, number

of approaches, and control

(uncontrolled, all-way stop,

two-way stop, yield,

signalized, roundabout). Ramp

termini are considered

intersections

• All MUTCD signs

• Barriers

• Median presence (Y/N), type

(depressed, raised, flush,

barrier)

• Rumble Strip presence (Y/N)

location (centerline, edgeline,

shoulder)

• Lighting presence(Y/N)

Types of Mobile Van Data

Existing roadway inventory data

acquired from agencies such as the

six State DOTs

(Data items not consistent)

• ~ 200, 000 centerline miles

• Includes HPMS files for the six

states plus:

• Functional Classification

• Signals

• Intersections

• Access Control

• Pavement Condition

• Bridge Location

• Vertical Alignment

• Interchanges

• Rest Areas

• Terrain

• Tunnels

• FRA grade crossings

Acquired Supplemental Data

Site

Total miles

collected % Rural/ Urban Routing purposes only

FL 4,366Rural: 45%

Urban: 55%

IN 4,635Rural: 64%

Urban: 36%

NC 4,558Rural: 59%

Urban: 41%

NY 3,570Rural: 68%

Urban: 32%

PA 3,670Rural: 83%

Urban: 17%

WA 4,277Rural: 31%

Urban: 69%

Total 25, 076

Existing data and information from State DOTs, Public Agencies, and Private Sources:

• ~ 200, 000 centerline miles • Crash history data• Traffic information – AADT• Traffic Data - continuous counts

(ATR)• Traffic Data -short duration counts• Aerial imagery • Speed limit data• Speed limit laws• Cell phone and text messaging

laws• Automated enforcement laws• Alcohol-impaired and drugged

drivers laws• Graduated driver licensing (GDL)

laws• State motor cycle helmet use laws• Seat belt use laws• Local climatological data (LCD)

NOAA• Cooperative weather observer/other

sources• Winter road conditions (DOT)• Work zone• 511 information• Changes to existing infrastructure

condition• Roadway capacity improvements

14

Data Summary

• 3,593 participants

• 3,958 vehicle-years

• 5 M trip files

• 35 M miles of driving

• 1,604 crashes, 2,778 near-crashes

• 12,538 centerline miles of roadway data collected by van

• 200,000 centerline miles of state roadway inventory data collected

15

SHRP 2 NDS Data Analyses

• 200 active or completed studies, 500 qualified researchers

• 68 published papers

16

NDS Study Topics

• Drivers

– Distraction: secondary tasks, cell phones

– Risk perception

– Teens; older drivers; drivers with ADHD etc.

– Belt use

– Urban driving profiles

• Pedestrians

– Signalized intersections

– High visibility crosswalks

• Speeding

– Speed limits

– Speeding and road geometry

17

Analysis Topics

• Roadways

– Rural 2-lane roads, horizontal and vertical curves, intersections

– Lane departures

– Freeway interchanges

– Offset left-turn lanes

• Vehicles

– Benefits of active safety and autonomous driving features

• Weather and lighting

– Roadway lighting

– Adverse weather conditions

• Work zones

• Modeling

– Calibrate traffic simulation models

Safer glances, driver inattention,

and crash risk in lead-vehicle

following

Trent Victor,

Jonas Bärgman, Christian-Nils Boda, Marco Dozza, Johan Engström, Carol

Flannagan, John D. Lee, Gustav Markkula

SHRP2 Phase 2 Final Report ETG Presentation 10 June, 2014

Research Topic

• Determine the relationship between driver

inattention and crash risk in lead-vehicle pre-crash

scenarios (rear-end crashes)

• Show which glance behaviors are safer than others

• Pinpoint the most dangerous glances away from the

road

Glance locations in

Random Baselines

Glance locations in

Matched Baselines

Glance locations before Near

Crashes (minTTC)

Glance locations before Crash

Rear-end crashes and near-crashes

• Most crashes result from a ”perfect mismatch” – the interaction

between last glance duration and the rate at which the situation

changed.

• Most crashes were associated with glances away from the road

shorter than 2 seconds.

• Brake lights have little effect on following driver reaction.

Implications

1. Distraction policy, regulation, guidelines: short glances can be risky.

2. Benefit of intelligent vehicle safety systems, e.g. Forward Collision

Warning.

3. Safety benefit of good highway operations: reduce sudden stops.

4. Teach safe glance behaviors: glance away from roadway only with

adequate headway.

How to Use the Data

25

26

Overview of the NDS Data

• Size: the file is huge

– 2 petabytes = 2 million 1 gig flash drives

– “Give me the whole raw data file” isn’t possible or sensible

• Complexity: different data types

– Categorical data constant over a trip: driver age, vehicle type

– Sampled data: collected at original resolution (once a trip up to 640

Hz during a crash): speed, acceleration, GPS position, radar,

vehicle network information

– Video data from 4 cameras; must be coded

• Automated reduction: lane tracker

• Manual reduction: all other items for specific analyses

• Privacy considerations: personally-identifying data (PII)

– Video and other personal information access only with IRB approval

for qualified researchers in secure location

27

Overview of the RID Data

• Size: the file is manageable

– 50-60 GB without video, 6-8 TB with video

• Complexity: 4 different data sources

– ESRI: baseline data for entire country, very few variables

– State roadway inventory data: from 6 study states; data vary by

state

– Mobile van data: very detailed, about 12,500 centerline miles;

includes forward video

– Supplemental data: from 6 study states, data vary by state

• Privacy considerations: should be none

– Video data may require IRB to determine exemption from IRB

review

Bite-sized NDS Data - How to Eat the Elephant

• Trip summary file

• Crashes, near-crashes, baseline files

– events and epochs

• Website data

• Other data enhancements

• Reduced data sets

28

29

Trip Summary File

• Trip summary file - categorical data on each trip(1 spreadsheet row per trip)

– Identify trips of interest

– Can be analyzed directly

• Variables

– Driver data – demographics, driver assessments

– Vehicle data – descriptive

– Roadway data – roadway class, speed limit, rural-urban

– Trip data – duration, speed, accelerations, headway, time to

collision, etc.

– Variables that change during a trip in bins, counts, or max/min:

• speed bins 0-10 mph, 10-20 mph, etc. – time or % of trip in each

• number of accelerations higher than threshold value

• minimum time to collision

30

Crashes, Near-Crashes, Baseline Files

• Crashes: 1,604, varying severity

– Most researchers want to examine crashes

• Near-crashes: “almost” crash but for sudden maneuver; 2,778

– Crash surrogates; how did driver avoid a crash

• Baseline: 20,000 random, 12,589 additional matched

– Denominator for risk calculations; measure overall prevalence

• Epoch files for each

– 30-second data segments (20 before, 10 after; only 20 for baseline)

– Includes most sensor data plus forward video

– Manual eye-glance coding, other coded variables

• Event files for each

– Categorical data coded from last 6 seconds of “before” data

– Manual video reduction; data dictionary on website

– Available on website

31

Website Data

• Data de-identified; no PII; fairly easy to access

• Descriptive data for whole data file

– Drivers – age and gender distributions, etc.

– Vehicles – type, age, etc.

– Trips – number, mileage, etc.

• Categorical data on all trips – from trip summary

• Event data from crashes, near-crashes, baseline

• Viewer for forward video and time series data display for crashes

and near-crashes – coming soon

– Approximately 20 seconds before precipitating event (e.g. lead

vehicle slamming on brakes) and 10 seconds after

339

34

Other Data Enhancements

• Radar data processing and coding

– Headway, time to collision

• Identify trips with possible alcohol use

– Passive alcohol sensor

• Cell phone records

– From cell phone providers or study participants

• Link NDS and RID files

– Identify all trips passing over a given roadway segment

– Identify all roadway segments over which a given trip travels

– Link matches trip IDs and roadway segment IDs

35

Reduced Data Sets

• Reduced data

– Trips or trip segments for specific research questions: trips with

teenage drivers; trip segments on rural 2-lane curves; …

– Retain only variables needed for research questions

For More Information

• Jim Hedlund, Highway Safety North

– jhedlund@sprynet.com

• David Plazak, TRB

– DPlazak@nas.edu

• InSight website (direct data access, information): https://insight.shrp2nds.us/

• TRB SHRP 2 Safety publications:

http://www.trb.org/Publications/PubsSHRP2ResearchReportsSafety.aspx

36

37

Questions and Discussion

Jim Hedlund

jhedlund@sprynet.com